Skip to content

Change needed in Whisper fine-tuning recipe to accommodate transformers4.30.0 #2016

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mravanelli merged 3 commits into from
Jul 5, 2023
Merged

Change needed in Whisper fine-tuning recipe to accommodate transformers4.30.0 #2016

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
dc1717f
changes required to fix in torch2.0
sangeet2020 Jun 6, 2023
cdda921
Fix with the solution used by Luca in the Benchmarks
mravanelli Jul 2, 2023
ce53199
Fix backward compatibility issues
lucadellalib Jul 2, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 10 additions & 1 deletion speechbrain/lobes/models/huggingface_whisper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,9 +103,18 @@ def __init__(
self._n_fft = feature_extractor.n_fft
self._hop_length = feature_extractor.hop_length
self._n_samples = feature_extractor.n_samples
# The following breaking changes were introduced in transformers>=4.29:
# 1) mel_filters.shape = (..., feature_extractor.feature_size) instead of (feature_extractor.feature_size, ...)
# 2) mel_filters.dtype = float64 instead of float32
# The following code fixes the issue in a backward compatible way
mel_filters = feature_extractor.mel_filters
if mel_filters.shape[0] != feature_extractor.feature_size:
mel_filters = mel_filters.T
assert mel_filters.shape[0] == feature_extractor.feature_size
self.register_buffer(
"_mel_filters", torch.as_tensor(feature_extractor.mel_filters)
"_mel_filters", torch.as_tensor(mel_filters, dtype=torch.float32)
)
#################################################################

self.model = WhisperModel.from_pretrained(source, cache_dir=save_path)

Expand Down